Enhanced outlier removal for extended Kalman filter based visual inertial odometry

Abstract

Outlier removal is a very important step for visual inertial odometry (VIO). Traditionally, outlier removal in extended Kalman filter (EKF) based VIO is achieved by Mahalanobis gating test. However, this simple test may not perform well for practical applications. One-point RANSAC is an effective approach for outlier removal. In this paper, we propose an enhanced approach based on one-point RANSAC. We employ feature re-projection error as an additional criterion to further identify outliers. Some experiments are conducted and the results are encouraging. The position and velocity deviation error of proposed method is better than that of the original one-point RANSAC algorithm.